If someone set up a custom GPT for your business in the last year or two, such as a quote-writer, a FAQ helper or a “turn my notes into a client email” assistant, this one’s for you.
OpenAI’s official migration FAQ confirms the planned move to plugins. When migration is available for your account, go to My GPTs → Migrate to plugin. Three details matter:
- Only the latest published version moves across. Finish essential edits first. The original becomes read-only after migration.
- Migrated plugins start private. Check installation and team access before switching.
- Custom actions do not transfer. Set up and test a replacement for any connection your workflow depends on.
Check your account notice. Some qualifying Enterprise and FedRAMP workspaces have a later retirement date. Migration availability varies. If someone else built the GPT, contact its creator about the replacement.
Start preparing now, so this can be a calm, planned swap rather than a scramble.
The calibration way to do it
I spent 20 years managing instrumentation at GM Holden’s Lang Lang Proving Ground. One rule never changed there.
The proving-ground rule
When an instrument was replaced with a new model, it didn’t go back into service until it read the same as the old one against a known standard.
Your custom GPT is an instrument, so treat it the same way. With AI, you’re checking that the result meets your standard—not that every word is identical.
One useful workflow
Set yourbaseline
- List every GPT your business uses.Note who built it, who uses it and what it connects to.
- Set your baseline before you migrate.Save three real inputs, along with outputs you were happy with. Add one trickier case to check the limits. Keep sensitive examples in your approved business systems.
- Migrate one GPT at a time.Changing one variable at a time is how you find what changed.
- Re-test against your baseline.Run the same inputs and compare accuracy, tone and format side by side. Check that the right reference material is used.
- Rebuild and test connected actions from start to finish.For example, send a clearly labelled test quote through to a test record in your CRM. Check the destination as well as the AI response.
- Re-share it on purpose.Confirm your team can use the replacement. My recommendation: keep human sign-off on customer-facing outputs for the first fortnight.
- Put a re-check in the calendar.Book it for a month’s time. Keep a short record of errors, fixes and the next test.
Why this matters more than it sounds
The workflow that breaks quietly is the one nobody remembers building. A GPT a contractor set up a year ago can sit inside your quoting process for months. If it stops working, you may find out from a customer before anyone on your team notices.
A short inventory now gives you time to plan. Allow more time for connected actions and team testing.
Your checklist this week
Ready to make the switch?
0 of 4 checked · For this visit only